2006 IEEE Ninth International Symposium on Spread Spectrum Techniques and Applications Fourier Transform Time Interleaving in OFDM Modulation Guido Stolfi and Luiz A. Baccalá Escola Politécnica - University of São Paulo Av. Prof. Luciano Gualberto, Trav 3, #158 São Paulo, SP, Brazil, 05508-900 Abstract This work introduces a new transformbased time interleaving algorithm: FTI-OFDM (Fourier Transform Interleaved OFDM) in which binary information is spread over several consecutive symbols that can be further scrambled in the frequency domain. Simulations are used to show its superiority over the usual binary time interleaving used in ordinary OFDM under several impairment scenarios that include impulsive noise and deep fading. Keywords Digital communications, OFDM, Time interleaving, digital television I. INTRODUCTION Orthogonal Frequency Division Multiplexing (OFDM) modulation has been widely used in communications systems due to its robustness against multipath distortion and fading. Application examples include the DVB-T and ISDB-T digital terrestrial television systems [1-3]. To achieve acceptable performance the insertion of both reference pilot carriers and intersymbol time guard intervals are necessary. Performance improvement under deep fading and impulsive noise is usually improved by employing long time interleaving (e.g. up to 0,5s in the ISDB-T system). However, time interleaving does not reduce total bit error rate; burst errors are spread in time until the average bit error rate is reduced within the capabilities of an error correcting code. This work introduces a novel transform-based time interleaving algorithm, herein called FTI-OFDM (Fourier Transform Interleaved OFDM) whereby the binary information is spread over several consecutive symbols by applying an inverse Fourier transform to the binary data. Comparative simulations between conventional OFDM and FTI-OFDM are presented for several channel impairments scenarios that include second-order effects arising from non-ideal channel estimation in mobile communications. This paper is organized as follows: after a short review of OFDM (Sec. II), the new algorithm is introduced in Sec. III, which is followed by schemes to overcome common impairments using FTI-OFDM (Sec. IV). Simulation results of several scenarios (Sec V) are followed by conclusions summarized in Sec. VI. II. BACKGROUND Figure 1 portrays a typical ordinary OFDM system whose baseband signal S(t) is formally described by: S t K 1 n0 k0 C( k) ( k, t) where C(k) is the complex data for the n-th OFDM symbol associated with by k-th carrier (k=0,...,k-1) where ( k, t) e 0 kk / 2 j 2 ( ttgnts) Tu nts t ( n 1) Ts t nts, ( n 1) Ts t represents each carrier, with Tg, Ts and Tu respectively standing for the guard interval, the total and the effective symbol durations (Ts = Tg+Tu). After data randomization and Forward Error Correction (FEC), binary data are grouped into symbols C(k) (one for each data carrier). Carriers (k,t) are usually modulated in QPSK, 16- or 64-QAM. An inverse Fourier transform is used to generate the modulated signal S(t) with symbol duration equal to Tu. A time guard interval is then generated, where a copy of the last samples of S(t), with duration Tg, are placed in front of the effective symbol data. This insertion is works against multipath distortion if Tg is longer than the associated propagation spread times of the received signal. Fig. 1 OFDM Modulation Pilot carriers, modulated with known amplitudes and phases, are intermixed with data carriers, and are used by the receiver to estimate and equalize the channel frequency response. This leads to reliable demodulation over fast 0-7803-9780-0/06/$20.00 2006 IEEE 158
changing environments (viz. for mobile and portable receivers). An important shortcoming of conventional OFDM is its sensitivity to very deep fades and to impulsive noise, in the form of wideband noise bursts, because the latter can affect all demodulated carriers in one or more consecutive symbols. To overcome this, typical OFDM systems employ long time interleaving to dilute such impairments. The interleave length is usually increased until the average error rate gets below the error correction capability [4]. III. THE PROPOSED METHOD In the present new proposal, conventional symbol time interleaving is replaced by an inverse Fourier transform performed on subsets of the digital data input. In this way, the individual carriers of an OFDM symbol are no longer modulated by discrete-amplitude QAM or PSK symbols, but rather by a non-quantized complex signal whose distribution is nearly gaussian. Figure 2 shows a block diagram of the proposed modulator (excluding input randomization and FEC and output guard interval insertion). The binary input b(m), previously randomized and FECencoded, is converted, p bits at a time, into symbols Q(k), using some complex modulation like QPSK or n-qam. For instance, p=6 for 64-QAM. A set of N symbols is then processed via an N-point inverse discrete Fourier transform, generating N complex samples C(k). These samples represent the amplitudes and phases that will modulate the k-th OFDM carrier, for n = 0 to N consecutive OFDM symbols. Transform of the OFDM generator. After IFFT transformatio the resulting M carrier symbols lead to the M time samples S(m) of the n-th transmitted symbol. Guard time insertion can be applied as usual on the resulting output signal S(t). The rationale behind the proposal is that the energy of impulsive noise occurring in a frame of N OFDM symbols is distributed among all N K symbols after demodulatio and produces constellation defocusing similarly to that produced by additive white gaussian noise of equal energy. Figure 3 illustrates the relationship among carriers, time and frequency in the FTI-OFDM modulation. In it the highlighted carrier symbols are the N samples that result from an inverse Fourier transform, performed on an input frame of N p bits. Frame synchronization at the receiver can be extracted from the pilot and auxiliary carriers (not shown). Fig. 3 Symbols and carriers in FTI-OFDM. For clarity, pilot and extra carriers are not shown. One should note that the present Fourier transform time interleaving scheme could also be used with single-carrier modulation systems. However, the peak-to-average ratio (PAR) of the modulated signal will be degraded in this situatio so its use is advisable only for multiple-carrier systems [5], where the transmitted signals already possess a near-gaussian distribution that is unaffected by FTI. Fig. 2 FTI-OFDM Modulation This process is repeated K times leading to a frame of N K samples C(k) which is then read row-wise, extracting K samples for some given n. These are then intermixed with M-K pilot, auxiliary data and null (guard band) carrier symbols, where M is the size of the output Inverse Fourier IV. FTI-OFDM: AVOIDING COMMON IMPAIRMENTS Fixed-frequency, narrowband interfering signals hamper all symbols associated with some carriers, leading to high error rates for those symbols. Narrowband energy interference can be spread over to all symbols by assigning different carriers to consecutive samples from the first IFFT (Fig. 2). This spread can be accomplished either by 159
randomization or via a simple carrier rotation scheme, such as: C (k) = C ( (k+n ) mod K) Figure 4 illustrates carrier rotation applied to an FTI- OFDM frame, in the hypothetical case of N = K. Data flow and processing can be made more uniform by skewing symbols to the carrier frames obtained from the first IFFT. In this way, as soon as an input frame of N p bits is obtained, one frame of carrier samples C(k) is generated and the first sample is made available for OFDM modulation. Figure 5 illustrates this scheme with and without carrier rotatio again with N = K. statistics etc.). When N = K, with time shifting, a single computational resource (IFFT engine) can be shared between interleaving and OFDM modulatio resulting in an efficient hardware implementatio as shown in Fig. 6. Fig. 6 Shared IFFT Implementation Fig. 4 Symbol Carrier Rotation V. SIMULATIONS The results presented here were obtained from simulating a baseband FTI-OFDM system, with N = K = 1024, p = 6 (64-QAM), with carrier rotation and no error correction encoding. Comparisons are made to ordinary OFDM with K=M = 1024 and 64-QAM modulation. The total number of encoded bits in the simulations is 6 1024 1024 (6.29 10 6 ). No time or frequency interleaving was used in ordinary OFDM, since total bit error rate is not affected. In all plots, except where noted, OFDM values are shown with cross symbols and FTI-OFDM with circles; the horizontal axis contains the SNR per bit (E b / N O ) while the vertical axis displays the observed average bit error rate. Fig. 5a Simple Time Shifting Fig. 5b -Time Shifting with Carrier Rotation Note that there is no restriction on the relative sizes of N and M; trade-offs are possible depending on transmission requirements (latency, computational resources, channel A. AWGN Channel FTI-OFDM performance for an AWGN channel is essentially equal to that of OFDM (Fig. 7). However, when the received signal is quantized to an 8-bit resolution (both real and imaginary axes), with the full signal scale set to +6 db above the mean signal power level (a typical configuration for digital signal processing), FTI-OFDM slightly outperforms ordinary OFDM. B. Random Impulse Noise Random, high-amplitude impulsive noise is particularly harmful to OFDM modulatio because it affects all carriers in a given symbol. In Fig. 8, randomly selected samples of S(t) suffered the addition of samples taken from a gaussian noise signal with the same mean power. The resulting BER is plotted against the ratio Rb, which is equal to the number of noisy corrupted samples divided by the total number of samples in a frame (1048576 samples). 160
error rate plotted in Fig. 11. The ratio Rf refers to the proportion of randomly deleted samples in an FTI frame. In this example, Fourier interleaving becomes advantageous when BER is less than 2 10-3. Fig.7 Comparative Bit Error Rate in an AWGN Channel Fig. 9 Effect of Narrowband Noise (Sec. V-C) Fig. 8 Comparative effect of Random Impulsive Noise as a function of the ratio Rb of noise corrupted samples (see Sec. V-B) OFDM FTI-OFDM Fig. 10 QAM Constellation With Narrowband Noise at the Receiver C. Narrowband Noise Since the total signal power is shared by many carriers, even low-level, fixed-frequency, narrowband interfering signals can destroy all data in all OFDM symbols within that frequency band. FTI-OFDM with carrier rotation spreads the interfering power over all carriers, reducing the total error rate except for very high interfering signal amplitudes. In Fig. 9, noise occupying a bandwidth equal to 1% of the total signal bandwidth was used in assessing the bit error rate. Fig. 10 compares 64-QAM OFDM constellations and the corresponding FTI-OFDM, both with a bit SNR of approx. 17 db. D. Deep Fading In this simulatio a block of consecutive samples was deleted (replaced by zero-valued samples) leading to the bit Fig.11 Bit Error Rate for Deep Fading 161
E. Signal Clipping FTI-OFDM is slightly more robust to signal clipping (before demodulation), as shown in Fig. 12, where the resultanting BER is plotted against the clipping ratio (in reference to the RMS signal amplitude). In the absence of noise, clipping at a peak level of 6 db above mean power (Clipping Ratio = 2) leads to a BER that is 5 times lower than conventional OFDM. Fig. 12 BER x Signal Clipping Level VI. CONCLUSIONS This paper presented a simple, robust and computationally efficient method for interleaving digital data in connection to OFDM modulation. This system can provide lower bit error rates for many channel impairments of interest when compared to conventional OFDM modulation. Current research focuses on the theoretical characterization of the new method. An FPGA prototype is being implemented for testing in connection with the Brazilian Digital TV System now under development. REFERENCES [1] ETSI EN 300 744, Digital Video Broadcasting (DVB): Framing structure, channel coding and modulation for digital terrestrial television, ETSI Draft EN 300 744 V1.4.1, 2001-01. [2] ARIB, Terrestrial Integrated Services Digital Broadcasting (ISDB- T): Specification of channel coding, framing structure and modulation, 1998. [3] Y. Wu, E. Pliszka, B. Caro P. Bouchard, G. Chouinard, Comparison of terrestrial DTV Transmission Systems: The ATSC 8-VSB, the DVB-T COFDM, and the ISDB-T BST-OFDM, IEEE Transactions on Broadcasting, Vol. 46, No. 2, June 2000 [4] H. Sari, G. Karam, I. Jeanclaude, Transmission Techniques for Digital Terrestrial TV Broadcasting, IEEE Communications Magazine, February 1995 [5] H. Ochiai, H. Imai, On the Distribution of the Peak-to-Average Power Ratio in OFDM Signals, IEEE Transactions on Communications, Vol. 49, No. 2, February 2001 162